Predicting prosthesis use and mobility needs in lower limb amputees: a machine learning approach using clinical and actigraphy data.

Only 53 participants were analysed, and 46 had prescribed K3, so the models may not represent K2 or K4 users. Validation was internal, testing on participants from the same sample rather than a new clinic, so real-world performance may be optimistic. The expert panel's reassessed K-level and goal decisions involved clinical judgment, and the weight given to each data source was not fixed. ActiGraph data did not capture terrain, environmental barriers, or reasons for not wearing the prosthesis. Attrition occurred before or early in monitoring, mainly because of in-person visit burden in the parent trial, so completers may differ from those who dropped out. Seasonal variation in activity may have affected model performance across different monitoring lengths.

The supplied text names the U.S. Department of Defense as the funder. It does not state author conflicts of interest or whether the funder designed the study.

Do not conclude that wearable actigraphy can replace clinical K-level assessment. The models predicting reassessed K-level were weak, with recall only 0.30 to 0.60, and the study did not test whether using these models improves patient outcomes.